The matching scores encoding the similarities between any two features are then refined using link PageRank and structure similarity to further enhance distinctive features while suppress noisy ones.
Combining the MODGDF and the spectral magnitude-based features gives a significant increase in recognition performance of 11% at best, while combining any two features derived from the spectral magnitude does not give any significant improvement.
For example, it may provide users with the feature types important to activity/identity, the correlation between features with high identity information, or the similarity between any two features.
Again, for a given selected feature (f_k), EFS-MI considers any two features say (f_i) and (f_j) as redundant, if (feature_feature_MI(f_i,f_k)=feature_feature_MI(f_j,f_k)) and exclude from further consideration.
Feature class relevance: It is defined as the degree of feature class mutual information between a feature f and a class label C. Feature-feature relevance: It is defined as the degree of feature-feature mutual information between any two features, say (f_i) and (f_j) where (f_i, f_j in F).
Given the set of importances and interactions defined above, we can easily verify that the outcome of the Choquet aggregation will exceed this threshold for combinations of any two features involving similarities for the phone number and at least one other feature (besides the date).
This is repeated exhaustively for all possible combinations constituting of any two features.
To qualify for features (measured by experiments) in our classifier, any two features must not be statistically correlated, must give a satisfactory distance separation in the feature space (between classes) and must offer good generalization for the predictor [ 110- 112].
Now, to classify an unknown sample using 2-feature model classifier, a combination of any two features and their corresponding pre-determined threshold values T is (selected from T i...all features for each dataset) were recruited.
